Flood forecasting with machine Learning, data Assimilation and Semi-pHysical modeling

Abstract : ANR FLASH project (2009-2013) intends to capitalize on the advantages of machine learning methods in order to provide tools for real-time flash floods forecasting. In a first step, water level forecasts were provided based on rain estimation of rainfalls, leading to the design of a demonstrating software. In a second step, weather RADAR measurements will be taken in advantage, as for rainfall estimation than for directly inputs reflectivity to the model. Comparison between the 3-type of inputs (rain gauge rainfall, RADAR rainfall, COMEPHORE reanalysis) will be assessed.